setup, which is described in section 4.2 in detail, was
used for measuring the two refrigerants in air. The polycarbonate Makrolon (Makrolon M2400, Bayer AG, Leverkusen Germany)
was used as sensitive layer. Details of the preparation of the 60 nm thick sensitive
layer can be found in .
to single analyte measurements with the sensor responses shown in figure
20 and in figure 22, two multicomponent
data sets were recorded , which are used for
the systematic investigation and development of different multivariate data
analysis methods. These data sets will be further referred to as "refrigerant
data". The first data set is a 21-level full factorial design whereby the
concentrations (relative saturation pressures) of the two analytes were varied
between 0 and 0.1 pi/pi0 relative saturation pressure.
This data set is used for the calibration and optimization of the different
multivariate models and will be further referred to as calibration data set.
The second data set is a 20-level full factorial design and will be referred
to as validation data set. The concentrations of this independent external validation
data set were varied between 0.0025 and 0.0975 pi/pi0
in 0.005 steps. Thus, all concentration levels of the two data sets are different
and consequently the validation data set should give a realistic estimate of
the network performance in a real world situation .
In total, 841 different mixtures of the refrigerants R22 and R134a were measured
by SPR. All measurements were performed in random order. The sensitive layer
was exposed to the analyte-air mixtures for 60 seconds and afterwards to synthetic
air for 300 seconds for recovery. During the sorption and desorption process
the signal was recorded with a resolution of 45 data points (see figure
22). The signals at each time point were used as independent variables whereby
only the 40 time points until 125 seconds of measurement time were used for
the multivariate data evaluation (see chapter 5 for
details of the time-resolved measurement proceedings).